Morphometric Analysis of Angiograms of Exudative Lesions in Age-RelatedMacular Degeneration

Objectives  To analyze the morphometric composition of lesion components in exudativeage-related macular degeneration and to study the relationships between individuallesion components and choroidal neovascularization (CNV) subtype, at 2 timepoints.

Methods  Morphometric analysis of 98 sets of angiograms separated by an intervalof at least 3 weeks, with no treatment delivered in the intervening periodbetween angiograms. Area measurements of individual lesion components weremade from digitally captured angiograms. Choroidal neovascularizations wereclassified into subtypes based on the proportions of classic CNV. Fully correcteddistance visual acuity measured on logMAR Early Treatment of Diabetic RetinopathyStudy charts was available at baseline and at a subsequent visit in 78 subjects.Data were analyzed using parametric and nonparametric tests, linear regression,and McNemar test of equal proportions.

Results  Wholly and predominantly classic CNVs were significantly smaller atinitial presentation than minimally classic or occult with no classic CNVs.Lesions containing blood and lipid were also significantly larger than lesionsnot exhibiting these features. Lesions containing any classic CNV expandedat a significantly greater rate than lesions without classic CNV. Approximately40% of lesions categorized as wholly classic CNV converted to predominantlyclassic CNV between baseline and the next follow-up visit.

Conclusion  The presence of classic leakage in exudative age-related macular degenerationis the most important risk factor for rapid expansion of CNV.

The development of choroidal neovascularization (CNV) is a devastatingcomplication of age-related macular degeneration (AMD) and leads to seriouslosses of central vision. Longitudinal cohort studies and the early randomizedcontrolled clinical trials undertaken by the Macular Photocoagulation StudyGroup have added to our understanding of the natural history of CNV expansion.^1-5 Thesestudies have determined that patterns of visual loss are affected by clinicaland angiographic features of the exudative macular lesion at presentation.^6,7 It is widely accepted that angiogramsperformed using a standardized protocol and subjected to a methodical gradingprocess yield robust information on visual and morphological outcomes. Thistype of analysis and interpretation of fluorescein angiograms has led to theclassification of CNV into distinct morphological subtypes.^3,8 Theclassification is based on the spatial and temporal characteristics of fluoresceinleakage, which is seen as hyperfluorescence, and the relative areas occupiedby the different lesion components.^3,8 Recentclinical trials vindicate such detailed classification, as treatment effectshave been shown to be dependent on the morphological composition of the CNV.^9-11 Some angiographicstudies^12,13 have documented growthrates in AMD lesions. In these studies, linear measurements on lesion diameterwere made, but analyses of the effect of lesion components on lesion growthand subclassification based on proportions of CNV were not undertaken.^12,13 Scrutiny of the literature yieldedno information on expansion rates of components of the neovascular lesionor the effect of lesion components on overall changes in lesion size.

The present study was therefore undertaken to analyze the morphometriccomposition of lesion components in exudative AMD at 2 time points. Furthermore,we examined the relationships between individual lesion components and CNVsubtype at these time points.

Angiograms were selected from an image database spanning January 1,1995, to June 30, 2001. Criteria for selection were as follows:

1. An angiographic diagnosis of exudative AMD withevidence of classic or occult (late leakage of indeterminate origin) CNV wasrequired.

2. Two angiograms a minimum of 3 weeks apart on thesame eye of any one patient were needed. If additional angiograms were available,a third was also selected for analysis.

3. Angiograms were required to have been performedaccording to a standardized protocol that included capture of stereocolorand angiographic frames on the selected eye, with late frames of both eyes.

4. No therapeutic intervention may have occurred inthe interval between angiograms.

When features suggesting chronicity (such as fibrosis involving ≥50%of the lesion or large lesions [≥6000 µm in greatest linear dimension])were seen on the first angiogram, these were considered exclusion criteria,as it was our clinical policy not to repeat angiography in such cases. Wealso did not include angiograms of eyes in which the primary lesion was afibrovascular pigment epithelial detachment, as our experience suggested that,when present, it often represented end-stage CNV. Most patients included inthe image database did not have indocyanine green angiography; therefore,we could not comment on the other phenotypes with features best recognizedon indocyanine green testing, such as idiopathic polypoidal choroidopathy.

Our target sample was sets of angiograms from 100 subjects. The angiogramsfrom 2 subjects were rejected after initial inclusion, because there was noleakage due to CNV in one and the lesion extended beyond the vascular arcadesin the other, which suggested chronicity.

Following selection of angiograms, the clinical notes were scrutinized,and if visual acuity had been measured within 1 week of the angiogram, thiswas entered into the database. Where visual acuity data were available, thishad been performed according to a standardized protocol using Early Treatmentof Diabetic Retinopathy Study logMAR charts under set conditions of illumination.

Fundus color and angiographic images were obtained using 2 digital capturesystems. Before April 2000, fundus photography and angiography were performedusing the Ophthalmic Imaging System (Ophthalmic Imaging Systems Inc, Sacramento,Calif) equipped with a Topcon SL7F fundus camera (Topcon, Surrey, Great Britain)and a black-and-white charge-coupled device camera (Sony Corp, New York, NY),with each image occupying 512 × 512 × 16-bit pixels, where 1 pixellength was equivalent to 17.6 µm. Accompanying color photographs werecaptured on 35-mm film. From April 2000 onward, images were captured on theTopcon retinal camera (model 50IX; Topcon) with IMAGEnet 2.11 software (IMAGEnet,Melbourne, Victoria). The high analogue definition charge-coupled device camera(HAD3CCD; Sony Corp), with 800 × 600-pixel resolution per chip for eachof the 3 colors, and a KodakMegaplus camera (model 1.4i/10; Redlake, Vianen,the Netherlands) were used for the capture of color and angiographic images,respectively. Images from angiograms captured on the Ophthalmic Imaging Systemwere imported into an Image Tools program (a free software program for imageanalysis) before grading. Images captured on the Topcon digital system wereanalyzed using IMAGEnet software.

Each angiogram, which generally consisted of some 40 frames, was viewedin its entirety on-screen. Selected angiographic frames were then magnifiedand viewed singly or as pairs using stereoviewers. Stereoviewing was necessaryto detect elevated lesion components. Total lesion area was defined as all CNV (classic or occult), plus any feature obscuringthe boundaries of the lesion and thus contiguous to the area of fluoresceinleakage. The definitions used to assign lesions to specific categories werebased on the proportions of classic CNV (Table 1) and conformed to the definitions previously described forgrading and analyzing angiograms.^3,8

Classic leakage was defined as clearly delineatedhyperfluorescence occurring early in the angiographic sequence, with blurringof the margins in the later frames. Late leakage of indeterminateorigin (a form of occult CNV) was defined as hyperfluorescence withindistinct margins, generally seen after 30 seconds had elapsed, which persistedinto late frames. Fibrovascular pigment epithelial detachment was said to be present when elevated speckled hyperfluorescence witha dome-shaped distribution was seen early in the angiographic series. Thepresence or absence of lipid exudation was also noted.

The planimetric area occupied by leakage from classic CNV and occultCNV (late leakage of indeterminate origin) was measured on selected angiographicframes in which these features were most clearly seen, by tracing around theborders of the hyperfluorescent region with the appropriate tool. The contributionsof blocked fluorescence due to blood and other components were distinguishedand the areas outlined, when present. When remotely present to the lesion,or when small specks of blood were noted within the lesion, area measurementswere not made. The size of the images was precalibrated in both systems, andarea measurements were generated in square millimeters. Other associated featuresof the CNV, such as subretinal fibrosis, atrophy, and subretinal fluid, wereestimated using a graded categorical approach, as the diffuse nature of thesecomponents often precluded detailed delineation of the edges. Therefore, thecontributions of each of these were estimated as accounting for less than25% of the lesion, 25% to 49%, 50% to 74%, or 75% or greater.

The position of the CNV with respect to the fovea was also recorded,and any change in position in the subsequent angiograms was noted. Two graders(F.A. and K.A.M.) who were trained in recognition of angiographic criteriaundertook the gradings. Grading was performed by both graders on every fifthangiogram to ensure interobserver reproducibility. All discrepancies in gradingwere resolved by one of us (U.C.).

All data were analyzed using the Statistical Packages for Social Sciences(SPSS Inc, Chicago, Ill), version 10. Summary statistics on CNV category atbaseline and subsequent visits and planimetric area measurements of lesioncomponents were generated. Only 38 subjects had been seen on 3 occasions;therefore, the final visit was designated as the first visit after baselineif only 2 visits had taken place, or the second visit if 3 visits had takenplace. To test for differences in lesion size by CNV subtype, we used thenonparametric Mann-Whitney test.

Linear regression analysis was used to examine the effect of baselinelesion components on change in lesion area at follow-up. A regression modelwas also used to analyze the effect of lesion components on change in distancevisual acuity. To facilitate comparison with the studies by Vander¹¹ and Klein¹² and coworkers,we examined lesion expansion in 58 eyes in which the interval between angiogramswas 120 days or less. The change in lesion diameter between baseline and thefirst visit was calculated, and an approximate linear growth rate per daywas estimated by CNV subtype.

To analyze longitudinal changes in CNV subtype, the baseline CNV categorywas cross-tabulated against the reassigned category at subsequent visits.We considered the change to be in the direction of occult CNV if the originalassignment showed greater proportions of classic CNV. Conversely, we consideredthe change to be in the direction of classic CNV if the original assignmentshowed a greater proportion of occult CNV. The significance of change wastested using the McNemar test of equal proportions.

Ninety-eight sets of angiograms from the 98 subjects were analyzed atbaseline and at the first visit, with 39 available from a second visit. Onlyone eye of any one subject was used as the study eye. The mean ± SDinterval between baseline and the first visit was 132 ± 84 days, andthat between the first and second visits was 169 ± 133 days. ExudativeAMD was present in 64% of fellow eyes and early-stage age-related maculopathyin the remainder.

The mean areas of the lesions at baseline and at the first visit byCNV subtype are shown in Table 2 and Table 3. When grouped by the presence orabsence of classic CNV, lesions containing any classic CNV were significantlysmaller at baseline and at the first visit than lesions without classic CNV.Similarly, when grouped by the presence or absence of occult CNV, lesionsin which occult CNV was present were significantly larger than those in whichoccult CNV was absent. Lesions that were composed of classic CNV only (100%classic) had the smallest mean area and were significantly smaller than lesionscomposed of 99% classic CNV or less. Lesions that consisted entirely of occultCNV (100% occult) were not significantly different from those lesions composedof 99% occult CNV or less.

When grouped by the presence or absence of blood at baseline, lesionswith blood were significantly larger at baseline (5.99 vs 3.25 mm², P = .01), the first visit (8.17 vs 5.52 mm², P = .04), and the final visit (8.56 vs 6.27 mm², P = .06). When the relationship at baseline between CNVsubtype and the presence of blood was examined, the mean ± SD proportionof classic CNV was higher in lesions without blood (blood present, 51.3% ±38.2% vs blood absent, 34.1% ± 31.2%; P =.02, independent samples t test). When grouped bythe presence or absence of lipid exudates, lesions with exudate were larger(5.55 vs 1.96 mm²; P = .08, nonparametricMann-Whitney test) and more likely to contain some occult CNV.

The mean lesion size increased from baseline to the first visit, asdid the mean area occupied by individual lesion components. A comparison ofthe mean lesion size at baseline and the first visit by lesion subtype atbaseline showed that increasing amounts of classic CNV generally resultedin proportionally greater expansion of the lesion (Figure 1). Figure 2 illustratesthe relationship between change in area of the lesion and classic and occultleakage with increasing interval size. The trend lines show that the changein area of classic CNV is steeper than that of occult CNV.

Table 3 shows the effectof lesion composition on lesion area. When grouped by the presence or absenceof classic CNV, the expansions in area of the lesion from baseline to thefirst visit (3.45 vs 1.60 mm², P = .03)and from baseline to the final visit (3.78 vs 2.15 mm², P = .04) were significantly different. When grouped by the presenceor absence of blood at baseline, eyes without blood did not exhibit significantlydifferent expansion rates compared with eyes with blood (data not shown).

Linear regression analysis, with change in area of the lesion at thefinal visit as the dependent variable, and with baseline lesion components,age, sex, and interval between angiograms as independent variables, showedthat the area of classic CNV at baseline significantly affected lesion expansion(t = 3.52, P<.001).

Table 4 shows the mean lesionarea at baseline and the first visit in the subgroup of eyes in which theinterval between angiograms was 120 days or less. The mean linear growth rateper day was 5.46 µm. When classic CNV was present, lesions grew faster,with a mean growth rate of 6.57 µm/d, compared with 3.92 µm/dwhen classic CNV was absent.

Regression analysis with change in distance visual acuity as the dependentvariable showed that the model that explained most of the variation includedchange in the area occupied by classic CNV at the final visit, distance visualacuity at baseline, and duration of follow-up.

The distribution of eyes by CNV category at baseline and the first visitis shown in Table 5. At the baselineassessment, only 12% of eyes were graded as having wholly classic CNV. Examinationof the distribution of eyes by CNV subtype at the first visit showed thatmost eyes remained in the category to which they were assigned at baseline.Conversions were noted in a small number of eyes, with 5 eyes with whollyclassic CNV at baseline having converted to predominantly classic CNV. Therewere no conversions from any of the other categories to wholly classic CNVduring the interval spanning baseline and the first visit. A change in categoryfrom predominantly classic to minimally classic or occult CNV, with no classicCNV between baseline and the first visit, occurred in a small proportion ofeyes. Three of 20 eyes initially graded as occult with no classic CNV haddeveloped a classic component, which led to a revised classification of minimallyclassic CNV at the first visit. We also examined the change in categorizationbetween baseline and the final visit (Table6). Most eyes remained in the categories to which they were assignedat baseline. Of the 24 changes noted, 12 were in the direction of increasingclassic CNV, while 12 were in the opposite direction (P = .99, McNemar test).

The present study used a systematic morphometric approach to quantifythe size of the exudative lesion and its constituents and made several importantobservations. We confirmed that wholly classic and predominantly classic CNVsare smaller at initial presentation than lesions with a lesser proportionof classic CNV. We found that the rate of expansion of classic CNV is fasterthan that of occult CNV, which would account for the more rapid increase inlesion size in the former.

Vander¹¹ and Klein¹² andcoworkers have reported mean daily growth rates for CNV membranes of 10 µmand 9 µm, respectively. Both of these studies used unidimensional measuresof expansion. The present study performed area measurements that detect changein all the boundaries of the lesion. To facilitate comparison between theseprevious studies and the present work, we estimated a linear growth rate oflesion expansion using angiograms in which the interval between angiogramswas 120 days or less. The growth rate of the lesion ranged from approximately6.5 µm/d when classic CNV was present to 3.9 µm/d when classicCNV was absent, with a mean growth rate of 5.5 µm/d. The slightly lowerrate of linear expansion in the present study may reflect differences in themethods used (we did not measure linear diameters) or the longer intervalsbetween angiograms in the present study, compared with that of Klein et al,in which the interval separating the angiograms was shorter, with a mean of13 days.

Our findings support previous observations that the presence of occultleakage is associated with larger lesion size. Recent clinical trials usingverteporfin photodynamic therapy attributed the treatment benefit seen ineyes with predominantly classic without occult CNV to the smaller size ofthe lesions in that subgroup.¹¹ The slowergrowth rate of the lesions in eyes containing occult CNV and a correspondingslower decline in visual function may permit such patients to remain eligiblefor entry into clinical trials, despite long-standing disease. However, thedata from the present study were from patients at first consultation. Therefore,the larger lesion size of occult CNV may simply reflect better levels of visualfunction in eyes with occult CNV than those with classic CNV, which wouldbe compatible with later presentation.

Although not reaching statistical significance (P = .08), our data suggest that the presence of lipid exudates was associatedwith larger lesion size and occult leakage. These findings are in accord withthe observations of Lozano-Rechy et al,¹³ whoreported that subretinal membranes in eyes with lipids tended to be ill defined.The present study suggests that blood is a risk factor for more rapid expansionof the lesion in eyes with occult CNV, which is consistent with the findingsof Stevens et al,¹⁴ in which variable growthrates in eyes with occult CNV were noted.

The present study demonstrated changes in lesion classification overtime. We accept that grader variability may account for some of the changesin lesion classification. However, the appearance of classic CNV when nonewas present or the conversion of wholly classic CNV into a predominantly classicsubtype is unlikely to be influenced by subjective factors. Furthermore, thepresent study found fewer than expected changes in classification of lesions.Notwithstanding a mean follow-up in excess of 4 months between baseline andthe first visit, only 13% of eyes developed a classic component when nonewas present at baseline. Stevens et al¹⁴ describedchanges in lesion size and lesion components in a subgroup of 40 subjectsenrolled in a preliminary randomized controlled trial of the macular gridlaser. Among 35 subjects with 9 months' follow-up, they found that 32% ofoccult lesions had doubled in size and that classic CNV had developed in 52%of eyes. The present study is not in complete accord with these previous observations,as we found that classic CNV only developed in some 15% of eyes that haveno classic CNV at initial presentation, nor did we find a doubling in sizeof occult lesions. The discordant findings suggest that the participant profilesin the 2 studies were different. The study by Stevens et al excluded subjectsif the amount of hemorrhage or blocked fluorescence was greater than the areaof visible CNV. The present study did not exclude such subjects, and in 13%of eyes, the amount of visible CNV was smaller than the area occupied by otherlesion components. The median change in distance visual acuity in the studyby Stevens et al was 2.5 lines. In the present study, the mean change in distancevisual acuity in eyes with occult with no classic CNV was 1.7 lines. It istherefore possible that the selection criteria used in the macular grid lasertrial resulted in the recruitment of a group of patients with a particularlyworse natural history, whereas the present study did not seek to exclude subjectson the basis of angiographic or visual acuity criteria. This may have resultedin the inclusion of angiograms from subjects with occult CNV without featuresthat predispose to rapid growth and vision loss.

A small number of eyes in which classic CNV accounted for 100% of thelesion were reclassified as predominantly classic or minimally classic CNVat a subsequent visit because of the appearance of occult leakage or otherlesion components. These findings emphasize the importance of early diagnosisand fast-tracking of patients with wholly classic or classic with no occultCNV, in whom visual outcomes have been shown to be improved with treatmentsuch as photodynamic therapy.

Corresponding author: Usha Chakravarthy, PhD, FRCS, FRCOphth, Departmentof Ophthalmology & Vision Science, School of Medicine, The Queen's Universityof Belfast, Belfast BT12 6BA, Northern Ireland, United Kingdom (e-mail: u.chakravarthy@qub.ac.uk).

Submitted for publication April 30, 2003; final revision received September23, 2003; accepted October 1, 2003.

This study was supported in part by strategic project grant UK G9404235from the Medical Research Council, London, England, and by a grant from theMacular Disease Society of the UK.

This study was presented in part as a poster at the 2002 Annual Meetingof the Association for Research in Vision and Ophthalmology; May 5, 2002;Fort Lauderdale, Fla.